Object detection device, object detection method, and program

1. An object detecting device that detects an object on a road surface included in stereo images, comprising:

a disparity map generator that generates a disparity map on the basis of the stereo images;

a road surface estimator that estimates a road surface region on the basis of the disparity map;

an object candidate location extractor that extracts pieces of disparity data above the road surface region from the disparity map, as an object candidate region where an object candidate is present;

an object identifying region extractor that extracts an object identifying region from the disparity map, the object identifying region including the object candidate region and having a width larger than the width of the object candidate region by a predetermined scale factor;

a geometric feature extractor that extracts a geometric feature in the object identifying region; and

an object identifying unit that identifies whether the object candidate is an object or a road surface on the basis of the geometric feature.

2. The object detecting device according to claim 1 , further comprising a table that associates the size and the disparity value of the object candidate region with the size of the object identifying region, wherein:

the object identifying region extractor extracts the object identifying region from the disparity map on the basis of the table and the size of the object candidate region.

3. The object detecting device according to claim 1 , wherein:

the object identifying region extractor calculates the size of the object candidate region in a real space on the basis of the disparity value and the size of the object candidate region,

determines whether the object candidate is a possible object that is present in a road environment on the basis of the calculated size of the object candidate region in the real space and preliminarily stored size information of possible objects that are present in the road environment in the real space,

extracts the object identifying region where the predetermined scale factor is a first scale factor when the object candidate is determined to be the possible object that is present in the road environment, and

extracts the object identifying region where the predetermined scale factor is a second scale factor when the object candidate is determined not to be the possible object that is present in the road environment.

4. The object detecting device according to claim 3 , wherein the first scale factor is smaller than the second scale factor.

5. The object detecting device according to claim 1 , wherein the lower end of the object identifying region is aligned with the lower end of the object candidate region.

6. The object detecting device according to claim 1 , wherein the predetermined scale factor is determined depending on the distance to the object candidate.

7. The object detecting device according to claim 1 , wherein the geometric feature extractor divides the object identifying region into a grid of squares to generate the geometric feature vector on the basis of representative values of the disparity data of each grid of squares.

8. The object detecting device according to claim 1 , wherein the geometric feature extractor divides the object identifying region into a grid of squares to generate the geometric feature vector on the basis of a correlation of representative values of the disparity data of each grid of squares.

9. The object detecting device according to claim 1 , wherein the object identifying unit performs identification using a result of learning based on a statistical learning.

10. The object detecting device according to claim 9 , wherein the object identifying unit performs the identification using a result of learning vertical displacements.

11. The object detecting device according to claim 1 , wherein the object identifying unit performs identification using the disparity value of zero when the object identifying region has parts where the disparity data is missing.

12. An object detecting method that detects an object on a road surface included in stereo images, comprising:

generating, by a disparity map generator, a disparity map on the basis of the stereo images;

estimating, a road surface estimator, a road surface region on the basis of the disparity map;

extracting, an object candidate location extractor, pieces of disparity data above the road surface region from the disparity map, as an object candidate region where an object candidate is present;

extracting, an object identifying region extractor, an object identifying region from the disparity map, the object identifying region including the object candidate region and having a width larger than the width of the object candidate region by a predetermined scale factor;

extracting, by a geometric feature extractor, a geometric feature in the object identifying region; and

identifying, by an object identifying unit, whether the object candidate is an object or a road surface on the basis of the geometric feature.

13. A computer executing a program for causing the computer to perform the steps of the object detecting method according to claim 12 .